Electromyogram Based Labor Prediction

Abstract

Human pregnancy is carrying a developing fetus within the female body which can be tested by many techniques. Pregnancy lasts for about nine months which is divided into three trimesters. Labor is when changes in anatomy and physiology occur in the female reproductive tract to prepare a fetus and placenta for delivery at the end of pregnancy. After the three stages of labor, birth will occur in two ways that is preterm/premature birth or term birth. Diagnosis of labor depends on the availability of uterine contraction and contraction monitoring devices that range from simple to complex electronics pressure sensors. But these monitoring/diagnosing devices are uncertain or they are applied for estimation of the date for term/preterm birth. The most common diagnosing device currently applicable is ultrasound, with an estimation date of 14 days range (the best estimation) to two months range (the worst estimation). Due to the poor accuracy of today’s maternal monitoring devices to diagnose labor and predict delivery, women admitted with the diagnosis of term or preterm labor are subsequently found not to be in true labor with misjudgments. If a wrong prediction of term/preterm is made by a physician, it makes many things difficult including tocolytic therapy, administration of steroids, and admission or transport to a hospital. The current study was able to demonstrate for the first time clinically that uterine electromyography (EMG) with age classification is another alternative to current human monitoring techniques. In addition, the riskiness of very young and old age pregnancy is demonstrated by considering maternal age as a major factor to predict term and preterm labor. In this work, the range of the estimation date has been reduced to one week. The research is implemented using an algorithm that utilizes a notch filter, Savitzky-Golay, and a band-pass Butterworth filter for preprocessing and wavelet transform for feature extraction. After feature extraction, three classification algorithms which are Support vector machine, Linear discriminant analysis, and Decision Tree were applied. The research used the Physio net database of labeled uterus EMG signals with different age levels at term or preterm labor. Using the wavelet transform, eight features were extracted and feed to the three classifiers. Two experiments were performed, age dependent and age independent classification. The overall accuracy attained were 88.78%, 100%, and 89.8% in the first experiment and 90.59%, 100% and 84.71% in the second experiment using Support vector machine, Linear discriminant analysis, and Decision Tree respectively. It was also found that the performances of the classifiers significantly depend on whether or no we take age of the pregnant into account.